U.S. Synchrotron Radiation Facility Achieves Intelligent Region Selection and Real-Time Image Reconstruction

Recently, the National Synchrotron Light Source II (NSLS-II) in the United States disclosed progress in intelligent X-ray imaging experiments: the research team integrated machine learning into the beamline experimental workflow, enabling automatic identification of regions within samples that warrant further measurement, and demonstrated real-time ptychographic reconstruction capability at 200 images per second. These achievements advance data analysis into the experimental process itself, providing a new technical pathway for improving the utilization efficiency of limited beam time.

In the automatic region selection experiment, the system acquires images while simultaneously analyzing complex samples, and then autonomously selects locations requiring high-resolution measurement. Researchers can therefore first conduct large-area scans and then concentrate fine measurements on regions of research value. This demonstration verified the feasibility of data-driven experiments, but does not yet mean that all samples and beamline experiments have achieved autonomous operation.

Another demonstration targeted three-dimensional imaging of microelectronic chips, accelerating ptychographic reconstruction through trained models and completing reconstruction of specific chip samples with approximately one-tenth the data volume of conventional requirements. Subsequent automatic segmentation further helped identify different materials and internal structures. These results were obtained under specific experimental conditions, and it cannot be directly assumed that other samples would achieve the same proportional data reduction.

The related work is part of the SYNAPS-I cross-laboratory collaboration, with the goal of further integrating measurement, reconstruction, and scientific analysis. The team's next step is to expand model training data and advance real-time three-dimensional imaging during experiments, so that images continuously improve as measurements proceed. This direction is expected to serve internal structure inspection of chips, research on degradation of battery materials, and characterization of complex nanomaterials, expanding the industrial application capabilities of large-scale synchrotron radiation scientific facilities.